badarg in eleveldb:iterator_move(Ref, prev) after 1.2.2p5

[uwiger]

I have yet to run this to ground, but using versions after 1.2.2p5, our test suites fail with a badarg when calling eleveldb:iterator_move(Ref, prev). Reverting back to 1.2.2p5 fixes the problem.

[uwiger]

The following test works with 1.2.2p5, but not with HEAD:

-module(ltest).

-export([t/0]).

t() ->
    os:cmd("rm -rf ltest"),  % NOTE
    {ok, Ref} = eleveldb:open("ltest", [{create_if_missing, true}]),
    try
    eleveldb:put(Ref, <<"a">>, <<"x">>, []),
    eleveldb:put(Ref, <<"b">>, <<"y">>, []),
    {ok, I} = eleveldb:iterator(Ref, []),
    {ok, <<"a">>, <<"x">>} = eleveldb:iterator_move(I, <<>>),
    {ok, <<"b">>, <<"y">>} = eleveldb:iterator_move(I, next),
    {ok, <<"a">>, <<"x">>} = eleveldb:iterator_move(I, prev)
    after
    eleveldb:close(Ref)
    end.

With current HEAD:

Eshell V5.9.2  (abort with ^G)
1> c(ltest).
{ok,ltest}
2> ltest:t().
** exception error: bad argument
     in function  eleveldb:async_iterator_move/3
        called as eleveldb:async_iterator_move(undefined,<<>>,prev)
     in call from eleveldb:iterator_move/2 (src/eleveldb.erl, line 188)
     in call from ltest:t/0 (ltest.erl, line 14)

Using bit bisect, I found that commit a465386 seemed to break iterators, but that commit gives another error:

Eshell V5.9.2  (abort with ^G)
1> ltest:t().
** exception error: no match of right hand side value {error,invalid_iterator}
     in function  ltest:t/0 (ltest.erl, line 14)

As reference, this is how it looks with 1.2.2p5:

Eshell V5.9.2  (abort with ^G)
1> c(ltest).
{ok,ltest}
2> ltest:t1().
{ok,<<"t:a">>,<<"x">>}
3> ltest:t2().
{ok,<<"t:a">>,<<"x">>}
4> ltest:t3().
{ok,<<"a">>,<<"x">>}

I'm running on a Mac, OSX 10.7.5, with OTP R15B02.

[matthewvon]

Prev() was intentionally left broken. The theory was that no one was using it and it would be one less thing to test. Do you actually use Prev(), or just have a test for it?

[uwiger]

Prev is crucial for us, so we must stay with 1.2.2p5 until it's fixed.

[matthewvon]

Ok. Once you have a position via Seek or SeekLast, do you typically call Prev() once or many times? Your answer will influence the nature of the Prev support.

[uwiger]

Calling it once is the more common scenario, but there are use cases where prev is called repeatedly.

[uwiger]

What is the status of this ticket? I just tried a recent version of eleveldb, and the badarg remains.

The use case I'm looking at currently iterates a user-specified number of times using prev.

[matthewvon]

Our unannounced decision was to address this after the 1.4 release. That release is not yet complete. So much code, so little time.

[jonmeredith]

But it can wait until after 1.4

On May 10, 2013, at 6:29 AM, Matthew Von-Maszewski notifications@github.com wrote:

Our unannounced decision was to address this after the 1.4 release. That release is not yet complete. So much code, so little time.

—
Reply to this email directly or view it on GitHub.

[matthewvon]

8 months later ... try branch mv-iterator-prev ... unit testing passes, will start bulk testing next.

[matthewvon]

bulk testing seems fine too. submitting a PR for the branch.

[bookshelfdave]

@uwiger do you mind if I include your code from above as an eunit test?

[uwiger]

Not at all - thanks for asking. :)